What GreptimeTeam/greptimedb shipped
Written by FoxPlug from the repository's own releases; not affiliated with Greptime. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/GreptimeTeam/greptimedb; Greptime did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of October 5, 2026
What shipped
- Nested MergeScan plans now execute on the datanode, laying groundwork for separately chosen broadcast-join plans. Pull request #9206
- The object store WAL now collects objects below a persisted reclaim boundary, adding garbage collection. Pull request #9471
- Object store WAL metrics and a conditional-create fault matrix were added. Pull request #9475
- OTLP trace ingestion skips repeated writes to service and operation names by caching them. Pull request #9432
- Import-v2 now batches logical table DDL through the logical-table batch DDL endpoint. Pull request #9467
- Instant queries of range-vector expressions no longer crash and return range-vector results. Pull request #9477
- The dashboard was updated to v0.14.0. Pull request #9510
- OTLP metric attributes with boolean values are converted into label values instead of being dropped. Pull request #9468
- TWCS assignment was aligned with half-open time windows so SSTs land in the intended window. Pull request #9464
- Memtables containing request-level skip-WAL writes are flushed on region close. Pull request #9440
Why it matters
This week brought fixes across query correctness, ingestion and storage behavior. PromQL instant queries of range-vector expressions no longer crash, boolean OTLP metric attributes no longer merge distinct data points, and TWCS window assignment was corrected. Storage and metadata side work also advanced, with object store WAL garbage collection and several metasrv reliability fixes.
Changelog entry
- Feat(datanode): execute nested MergeScan plans on the datanode Pull request #9206
- Feat(log-store): collect object store WAL objects below a persisted reclaim boundary Pull request #9471
- Feat(log-store): add object store WAL metrics and the conditional-create fault matrix Pull request #9475
- Perf: cache OTLP trace service and operation writes Pull request #9432
- Feat: batch logical table DDL during import-v2 Pull request #9467
- Fix(promql): return range-vector results for instant queries Pull request #9477
- Fix(otlp): convert boolean metric attributes into label values Pull request #9468
- Fix: align TWCS assignment with half-open time windows Pull request #9464
- Fix(mito): flush memtables containing skip-WAL writes on region close Pull request #9440
- Feat: update dashboard to v0.14.0 Pull request #9510
- Fix(promql): keep vector labels when a computed scalar is on the right Pull request #9478
- Fix(metasrv): retry failed heartbeat address updates Pull request #9437
This week: instant queries of range-vector expressions no longer crash, boolean OTLP metric attributes are kept as labels, TWCS windows were aligned, and object store WAL garbage collection landed.
This week in GreptimeDB: instant queries of range-vector expressions now return results instead of crashing, boolean OTLP metric attributes are converted into labels rather than dropped, and TWCS assignment now follows half-open time windows. On the storage side, the object store WAL gained garbage collection below a persisted reclaim boundary plus metrics and a conditional-create fault matrix. OTLP trace ingestion now caches service and operation writes, import-v2 batches logical table DDL, and nested MergeScan plans run on the datanode.
Week of September 28, 2026
What shipped
- Released v1.3.0-beta.1 with native histogram ingestion enabled by default for Prometheus Remote Write v2. Release
- Implemented PromQL @ modifier on vector and matrix selectors for timestamp-based evaluation. Pull request #9224
- Added CORS support for gRPC-Web on the frontend gRPC server to enable browser clients. Pull request #9374
- Added HDFS object storage backend as a datanode storage option for locally deployed HDFS. Pull request #8701
- Wired object store WAL into datanode and standalone server for durability. Pull request #9386
- Enabled automatic column addition for Flight bulk inserts to support schema-on-write handling. Pull request #9285
- Allowed customized time index unit for metric engine tables beyond millisecond precision. Pull request #9236
- Added WAL durability wait before publishing manifest watermark for cross-region recovery. Pull request #9385
- Reduced allocations when building bloom and inverted indexes during compaction. Pull request #9359
- Optimized catalog
batch_getto skip backend calls on full cache hits. Pull request #9431
Why it matters
V1.3.0-beta.1 brings production-ready features for Prometheus compatibility and query evaluation, while infrastructure improvements to WAL handling, storage backends, and index performance address durability and scale concerns. The week focused on stability improvements and finishing components needed for cross-region deployments.
Changelog entry
- Native histogram ingestion enabled by default for Prometheus Remote Write v2 Release
- PromQL @ modifier implemented on vector and matrix selectors Pull request #9224
- CORS support added for gRPC-Web on frontend gRPC server Pull request #9374
- HDFS object storage backend added for datanode deployments Pull request #8701
- Object store WAL wired into datanode and standalone server Pull request #9386
- Automatic column addition supported for Flight bulk inserts Pull request #9285
- Customized time index units now supported for metric engine tables Pull request #9236
- WAL durability wait added before manifest watermark publishing Pull request #9385
- Allocation overhead reduced for bloom and inverted index building Pull request #9359
- Bloom filter searches batched across row groups for improved performance Pull request #9361
- Catalog
batch_getoptimized to skip backend calls on full cache hits Pull request #9431 - V2 series scan temporarily disabled by default pending fixes Pull request #9436
GreptimeDB v1.3.0-beta.1 is out with native histogram support, PromQL @ modifier, gRPC-Web CORS, HDFS storage, and WAL durability guarantees. Production release coming soon.
GreptimeDB v1.3.0-beta.1 brings major improvements for production deployments: native Prometheus histograms enabled by default, PromQL @ modifier for timestamp-based queries, CORS support for browser clients, HDFS storage backend, and WAL durability guarantees for cross-region recovery. The release focuses on Prometheus compatibility and infrastructure reliability.
Week of September 21, 2026
What shipped
- Added object store WAL durable write path for log-store. Pull request #9320
- Implemented chaining of object store WAL objects with recovery of the latest chain. Pull request #9334
- Added manual series index reconciliation feature. Pull request #9323
- Fixed vector aggregates to work correctly with GROUP BY and partial aggregation. Pull request #9338
- Fixed PromQL vector matching to derive result labels correctly and reject ambiguous matchings. Pull request #9306
- Enabled native histogram ingestion by default for Prometheus Remote Write v2. Pull request #9301
- Switched to stable Rust 1.96.1 and removed all nightly feature gates. Pull request #9303
- Added pending rows batching support for MySQL and PostgreSQL. Pull request #9302
- Optimized performance by lazily decoding dense primary key columns. Pull request #9226
Why it matters
This week focused on WAL durability improvements, query correctness fixes for aggregates and PromQL matching, and toolchain stabilization. The combination of object store WAL chaining, histogram ingestion, and vector matching fixes improves both reliability and query accuracy for users.
Changelog entry
- Log-store: add object store WAL durable write path Pull request #9320
- Log-store: chain object store WAL objects and recover latest chain Pull request #9334
- Series index: add manual reconciliation Pull request #9323
- Query: fix vector aggregates with GROUP BY and partial aggregation Pull request #9338
- Promql: derive vector matching result labels and reject ambiguous matchings Pull request #9306
- Otlp: enable native histogram ingestion by default Pull request #9301
- Otlp: add histogram rejection metrics and compact zero buckets Pull request #9321
- Toolchain: switch to stable Rust 1.96.1 Pull request #9303
- MySQL/PostgreSQL: support pending rows batching Pull request #9302
- Mito2: lazy decode dense primary key columns Pull request #9226
- Promql: push label filters into grouped join inputs Pull request #9280
- Mito: limit approximate series index disk usage Pull request #9313
- SchemaManager: add schemas metadata stream Pull request #9325
- Flow: freeze recovery windows and retention bounds for incremental flows Pull request #9312
- Json2: support altering JSON2 settings Pull request #9029
- Removed experimental vector index implementation Pull request #9345
This week: object store WAL chains, vector aggregate fixes, PromQL matching corrections, native histogram support enabled, and stable Rust 1.96.1. Several query correctness and performance improvements shipped.
This week's changes strengthen GreptimeDB's durability and query correctness. We implemented object store WAL chaining with recovery, fixed vector aggregate and PromQL vector matching issues, enabled native histogram ingestion by default, and switched to stable Rust toolchain. Performance improvements include lazy decoding of primary keys and better label filter pushing in PromQL joins.
Week of September 14, 2026
What shipped
- Trace V2 ingestion now available with JSON2 attributes for span, scope, and resource data. Pull request #9192
- Jaeger queries now supported for Trace V2 with optimized writes. Pull request #9257
- HTTP protocol now batches ordinary table writes for improved efficiency. Pull request #9115
- Two-phase series scans now use series and range indexes to improve performance. Pull request #9153
- Range index reads and builds made opt-in via
experimental_enable_range_indexflag. Pull request #9219 - Object store WAL log store now supports construction and recovery. Pull request #9238
- Region statistics now expose min and max timestamps. Pull request #9060
- PromQL binary operands now propagate matching-label filters for better filtering. Pull request #9202
- Prometheus remote read now merges duplicate timeseries across record batches. Pull request #9032
- Primary key range comparison across schema versions fixed in mito2 compaction. Pull request #9205
Why it matters
This week brings Trace V2 fully online with JSON storage and Jaeger query support, enabling better observability data handling. Performance work across PromQL, series scanning, and HTTP batching reduces resource usage. Infrastructure improvements to object store WAL and region statistics support future scaling.
Changelog entry
- Feat(otlp): add trace v2 ingestion with JSON2 attributes Pull request #9192
- Feat(trace): support Jaeger queries for Trace V2 and optimize writes Pull request #9257
- Feat: batch ordinary table writes across HTTP protocols Pull request #9115
- Perf(mito2): use series and range indexes in two-phase series scans Pull request #9153
- Feat(mito2): make range index reads and builds opt-in Pull request #9219
- Perf(promql): propagate matching-label filters between binary operands Pull request #9202
- Fix(servers): merge duplicate timeseries across RecordBatches in Prometheus remote read Pull request #9032
- Feat(log-store): add object store WAL store construction and recovery Pull request #9238
- Feat: expose region min/max timestamp in
region_statisticsPull request #9060 - Feat(otlp): preserve trace v2 events and links as JSON Pull request #9232
- Fix(mito2): compare primary key ranges across schema versions Pull request #9205
- Feat(mito2): split SWCS output files by size threshold Pull request #9259
- Perf(mito2): lazily materialize sparse primary key tags Pull request #9199
- Perf(servers): reduce temporary memory in protocol handling Pull request #9249
- Feat(json2): respect type hints during JSON2 type concretization Pull request #9222
- Perf(otlp): share trace resource and scope attributes Pull request #9253
- Feat: add status label to datanode failed-insert metric Pull request #9191
Trace V2 ingestion is live with JSON2 attributes. Jaeger queries now work with Trace V2. HTTP writes batch ordinary tables. Performance optimizations in PromQL, series scans, and mito2.
This week's GreptimeDB release brings Trace V2 to production with full JSON attribute support and Jaeger query compatibility. We optimized HTTP write batching for ordinary tables, improved PromQL filter propagation, and enhanced two-phase series scans with index usage. Object store WAL infrastructure for log stores is now complete, and region statistics now expose timestamp bounds. Configuration flexibility improved with opt-in range index support.